TimeLogic Challenge @ CVPR 2026: Strong MLLMs Meet Evidence-Seeking Agents for Temporal-Logic Video Question Answering

2026-05-31

AI Summary

This work addresses the limitations of conventional fixed-frame sampling in video temporal reasoning for question answering, which often fails to capture localized or dispersed temporal evidence. To overcome this, the authors propose a training-free active exploration framework that models question answering as a Think-Act-Observe loop. The framework integrates multi-granularity video sampling, a timestamp alignment mechanism, and explicit temporal reasoning to perform quantitative analysis on a unified timeline. Guided by the question category, the system dynamically routes to tailored strategies, adaptively adjusting its sampling budget and reasoning depth. Built upon Gemini 1.5 Pro, it incorporates lightweight classifiers and category-specific prompts. Evaluated on the TimeLogic benchmark, the method achieves an average accuracy of 77.13%, demonstrating substantial improvements in understanding complex temporal relationships.

Abstract

Temporal-logic video question answering requires a model to reason about when actions occur relative to one another, such as before, after, until, since, overlap, and multi-event chains, rather than merely what is present in a video. Standard vision-language models typically answer such questions in a single pass over a fixed, uniformly sampled set of frames, which is poorly matched to evidence that is often localized to narrow action boundaries or dispersed across several distant events. We present an evidence-seeking agent that treats temporal-logic VideoQA as active exploration. The agent follows a Think-Act-Observe loop driven by a multi-granular sampling toolkit, where every observation is interleaved with its absolute timestamp so that temporal relations reduce to numerical comparisons on a shared time axis. Its behavior is shaped by benchmark structure: a lightweight classifier routes each question to a temporal category, each with a tailored policy, iteration depth, and prompt, while sampling budgets adapt to corpus characteristics and clip length. The resulting training-free system couples Gemini 3.1 Pro with a temporal-reasoning policy and achieves 77.13 AvgAcc on the official TimeLogic test set.

Problem

• Temporal-logic VideoQA
• time reasoning
• evidence localization
• multi-event chains
• temporal relations

Innovation

• evidence-seeking agent
• temporal-logic reasoning
• active exploration
• multi-granular sampling
• timestamp-aware observation